The AI Field Guide / I

Letter I

4 terms, explained without the techno-murk.

/

Indigenous data

Everyday

Data connected to Indigenous peoples, communities, lands, cultures, resources or knowledge.

This can include health records and population statistics as well as language, genealogy, traditional knowledge and environmental information. Indigenous data sovereignty holds that Indigenous peoples have rights and authority concerning how such data is collected, controlled, interpreted and used.

For example

A research team works with an Indigenous community to agree who may access cultural knowledge and how any AI system may use it.

#

Inference

Everyday

Using a trained model to produce an answer or prediction.

Training is the learning stage; inference is the doing stage. Every time a model classifies an image, predicts a number or replies to a prompt, it is running inference.

For example

When you ask a chatbot a question and it replies, that reply is produced during inference.

#

Input

Start here

The information given to an AI system.

An input might be a prompt, image, voice recording, document, database row or sensor reading. The quality and clarity of the input affect the result.

For example

A photograph uploaded for description is an input.

#

Instruction tuning

Deeper

Training that teaches a model to respond usefully to human instructions.

A raw pretrained model mainly predicts what text comes next. Instruction tuning gives it examples of requests and good responses so it behaves more like an assistant.

For example

The model learns that 'summarise this' should produce a concise summary rather than continue the source text.

#